COMPARING RESULTS OF EUCLIDIAN DISTANCE & HMM FOR SINGER IDENTIFICATION

Abstract — Singing voice identification is different from speaker recognition as there are significant differences between singing voice & speech in terms of production & perception. To identify singer, there are two main steps, feature extraction & classification. Generally used feature extraction methods are LPC, MFCC with classifiers such as HMM, GMM, SVM & ANN.In this paper we propose system for singer identification using MFCC as feature extractor & HMM as classifier. MFCC can capture the acoustic characteristics for speech recognition, speaker recognition, and other speech related applications easily they extract a lot of speaker-independent information. The process starts with extraction of features from audio signal then it will generate feature vectors these are given to classifier for further classification the HMM used as classifier which is best suited for singing voice recognition as it represents hidden as well as observed states HMM is Probabilistic model for classification. In this proposed method, the pre emphasis is processes before MFCC calculation for compensate the frequencies that are suppressed during sound production. An experimental result in this paper shows that it gives about 75 % accuracy for Euclidian Distance & 31.5 % for HMM classifier for the given set of data